Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/xcota/pos/recallnpx skills add xcota/pos --skill recallgit clone --depth 1 https://github.com/xcota/posWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/xcota/pos/recall)<a href="https://agentmods.dev/skills/xcota/pos/recall"><img src="https://agentmods.dev/badge/skills/xcota/pos/recall.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00016 | $0.00543 |
| Opus 5 | $0.00008 | $0.00271 |
| Sonnet 5 | $0.00003 | $0.00109 |
| Haiku 4.5 | $0.00002 | $0.00054 |
Grade A, and why
recall scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
/recall — Semantic Memory Search
/recall {query} — top relevant nodes (file + section + snippet) from the node-level index.
Mechanism
scripts/memory_index.py — EmbeddingGemma q8 ONNX, nodes chunked by header, one .npz sidecar (state/memory-index/), brute-force cosine. Zero DB, zero external APIs. A warm server on 127.0.0.1:8765 (lazy, 30 min TTL): the first query is ~7s, then ~65 ms. Architecture: knowledge/concepts/memory-embedding-layer.md.
Setup (once): the engine runs in a venv. If search errors with a missing module, run bash scripts/memory_index_setup.sh, then .memory_venv/bin/python scripts/memory_index.py build (use the venv python, NOT system python3 — the deps live in the venv). First run downloads a ~309 MB model; build is seconds on a small vault, up to ~35 min on a large one.
Steps
- Run the search from the vault root:
(It knocks on the warm server itself; no server → a cold search + a lazy background spawn. If you named the venv differently, use that path — e.g..memory_venv/bin/python scripts/memory_index.py search "$QUERY"bge_test_venv/bin/python.) - Result format:
rank. [score] relpath+§ header-path+ snippet. Score ≥0.5 = usually an exact hit; 0.4–0.5 = eyeball it. - The result is a set of Read candidates, NOT the answer. Open the top 1–3 files and read the section.
- No sensible results → rephrase the query (semantics prefer a question, not a keyword stub).
- Do NOT synthesize — that's the caller's job.
When to call
- Main thread doesn't remember context; "did we already discuss this?"
- A meaning-level question with no exact term (jargon, paraphrase)
- A subagent needs historical context
When NOT to call
- Exact lookup of a known file → Read
- Exact term/name → Glob/grep (faster; semantics get appended by the hook if it's enabled)
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago First seen · 37 lines · 16 tokens per session scan A 4dd128a6a13b
recall is a skill published in the GitHub repository xcota/pos (43 stars, last pushed 1mo ago), licensed MIT. It adds 16 tokens to every session and 543 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
ori-memory
Persistent agent memory with learning retrieval. Knowledge graph on markdown files — capture insights, decisions, research, and learnings during work, then retrieve them weeks or months later. Use when knowledge is too valuable to lose but too much to inject into every prompt.
capture-knowledge
将对话、项目文档、调试经验或用户指定材料沉淀到 Second Brain。用于“保存知识”“提取经验”“导入文档”“记录到知识库”“capture knowledge”等请求；先保存可追溯来源，再按一个稳定主题一张卡提取或更新 canonical 页面，禁止把 README、报告、故事或整段会话直接做成一张大卡。.
second-brain-attach
把全新、重装或重置后的 AI Agent 接入用户本地 Second Brain：安装七个 Second Brain Skills，连接薄 MCP，注入通用规则与该 Agent 的独立身份，并在全新会话中验证。用户要求安装、接入、迁移、初始化、恢复 Codex、Claude、AntiGravity、Hermes 或兼容 Agent，或者希望一个新 Agent‘变成我的形状’时使用，即使用户没有说出本 Skill 名称。.
second-brain-distill
把选定的历史 AI Agent 对话完整榨取为可长期保存的规则、知识、Agent 配置、Skills、项目经验、山脉证据和晋升附件；只有恢复性与完整性门禁通过后，才删除明确 allowlist 中的原始批次。用户要求挖掘、榨取、归档、压缩、清理或删除旧 Agent 对话，同时保留其中全部有价值信息，或提到历史对话数据挖掘、SkillOpt 理念时使用。.
second-brain-doctor
诊断一个 Agent 是否正确接入用户的 Second Brain：检查权威文件、通用与 Agent 专属投影、全部七个 Second Brain Skills、薄 MCP 连通性与 allowlist、local-only 或 private-remote Git 边界、漂移、过时旧 owner 和全新会话行为。用户询问是否健康、装好、连接、同步、可迁移或正常工作，完成 Attach/Sync 后，或 Agent 行为不符合既有偏好时使用。默认只读。.
second-brain-learn
从当前 Agent 对话和工作轨迹中挖掘可长期保留的偏好、方法、Agent 角色事实、项目规则、知识与山脉进展；审计现有 owner，新增、纠错、缩限、替代、合并、迁移或退役规则，自动合并有证据的低风险变化，高风险变更先询问用户，并同步当前 Agent。用户说学习、记住、采纳或停止某种行为、修正规则、保存本次经验、从当前对话改进、运行 /learn，或要求未来 Agent 遵守某项教训时使用。.